Removed API 1.0 from NVIDIA plugin (openvinotoolkit#835)

* Removed API 1.0 from NVIDIA plugin * Fixed chacing tests
sunxiaoxia2022 · Jan 18, 2024 · a16af94 · a16af94
1 parent 1c448b2
commit a16af94
Show file tree

Hide file tree

Showing 27 changed files with 23 additions and 156 deletions.
diff --git a/modules/nvidia_plugin/include/nvidia/nvidia_config.hpp b/modules/nvidia_plugin/include/nvidia/nvidia_config.hpp
diff --git a/modules/nvidia_plugin/src/cancellation_token.hpp b/modules/nvidia_plugin/src/cancellation_token.hpp
@@ -4,8 +4,6 @@
 
 #pragma once
 
-#include <ie_extension.h>
-
 #include <atomic>
 #include <error.hpp>
 #include <functional>

diff --git a/modules/nvidia_plugin/src/cuda/blas.hpp b/modules/nvidia_plugin/src/cuda/blas.hpp
@@ -5,7 +5,6 @@
 #pragma once
 
 #include <cublas_v2.h>
-#include <ie_extension.h>
 
 #include "runtime.hpp"
 

diff --git a/modules/nvidia_plugin/src/cuda_compiled_model.cpp b/modules/nvidia_plugin/src/cuda_compiled_model.cpp
@@ -19,7 +19,6 @@
 #include "memory_manager/cuda_immutable_memory_block_builder.hpp"
 #include "memory_manager/cuda_memory_manager.hpp"
 #include "memory_manager/model/cuda_memory_model_builder.hpp"
-#include "nvidia/nvidia_config.hpp"
 #include "nvidia/properties.hpp"
 #include "openvino/runtime/exec_model_info.hpp"
 #include "openvino/runtime/internal_properties.hpp"

diff --git a/modules/nvidia_plugin/src/cuda_compiled_model.hpp b/modules/nvidia_plugin/src/cuda_compiled_model.hpp
@@ -32,7 +32,7 @@ class CompiledModel : public ov::ICompiledModel {
                   const Configuration& cfg,
                   const std::shared_ptr<ov::threading::ITaskExecutor>& wait_executor,
                   const std::shared_ptr<const ov::IPlugin>& plugin,
-                  bool loaded_from_cache = false);
+                  bool loaded_from_cache);
 
     ~CompiledModel();
 

diff --git a/modules/nvidia_plugin/src/cuda_config.cpp b/modules/nvidia_plugin/src/cuda_config.cpp
@@ -3,10 +3,10 @@
 //
 
 #include "cuda_config.hpp"
+#include "openvino/runtime/internal_properties.hpp"
 
 #include <fmt/format.h>
 
-#include <cpp_interfaces/interface/ie_internal_plugin_config.hpp>
 #include <error.hpp>
 #include <regex>
 
@@ -154,19 +154,6 @@ Configuration::Configuration(const ov::AnyMap& config, const Configuration& defa
 
         if (ov::num_streams == key) {
             num_streams = value.as<ov::streams::Num>();
-        } if (NVIDIA_CONFIG_KEY(THROUGHPUT_STREAMS) == key) {
-            if (value != NVIDIA_CONFIG_VALUE(THROUGHPUT_AUTO)) {
-                try {
-                    num_streams = value.as<ov::streams::Num>();
-                } catch (...) {
-                    throw_ov_exception(
-                        fmt::format("NVIDIA_CONFIG_KEY(THROUGHPUT_STREAMS) = {} "
-                                    "is not a number !!",
-                                    value.as<std::string>()));
-                }
-            } else {
-                num_streams = ov::streams::AUTO;
-            }
         } else if (ov::device::id == key) {
             // Device id is updated already
             continue;
@@ -215,11 +202,6 @@ ov::Any Configuration::get(const std::string& name) const {
     } else if (name == ov::num_streams) {
         return (num_streams == 0) ?
             ov::streams::Num(get_optimal_number_of_streams()) : num_streams;
-    } else if (name == NVIDIA_CONFIG_KEY(THROUGHPUT_STREAMS)) {
-        auto value = (num_streams == 0) ?
-            ov::streams::Num(get_optimal_number_of_streams()) : num_streams;
-        return (value ==  ov::streams::AUTO) ? NVIDIA_CONFIG_VALUE(THROUGHPUT_AUTO)
-                                             : ov::util::to_string(value);
     } else if (name == ov::hint::num_requests) {
         return hint_num_requests;
     } else if (name == ov::hint::inference_precision) {

diff --git a/modules/nvidia_plugin/src/cuda_config.hpp b/modules/nvidia_plugin/src/cuda_config.hpp
@@ -8,7 +8,6 @@
 #include <memory>
 #include <string>
 
-#include "nvidia/nvidia_config.hpp"
 #include "openvino/runtime/properties.hpp"
 #include "openvino/runtime/threading/istreams_executor.hpp"
 

diff --git a/modules/nvidia_plugin/src/cuda_infer_request.cpp b/modules/nvidia_plugin/src/cuda_infer_request.cpp
@@ -8,7 +8,6 @@
 #include <fmt/format.h>
 
 #include <algorithm>
-#include <description_buffer.hpp>
 #include <gsl/span_ext>
 #include <map>
 #include <memory>

diff --git a/modules/nvidia_plugin/src/cuda_operation_base.cpp b/modules/nvidia_plugin/src/cuda_operation_base.cpp
@@ -5,6 +5,8 @@
 #include <utility>
 
 #include "cuda_operation_base.hpp"
+#include "openvino/core/node.hpp"
+
 namespace ov {
 namespace nvidia_gpu {
 
@@ -14,8 +16,8 @@ OperationBase::OperationBase(const CreationContext& /*context*/,
                              IndexCollection&& outputIds)
     : node_name_{node.get_friendly_name()},
       type_name_{node.get_type_info().name},
-      input_ids_{move(inputIds)},
-      output_ids_{move(outputIds)} {
+      input_ids_{inputIds},
+      output_ids_{outputIds} {
     if (node.get_input_size() > 0) {
         runtime_precision_ = node.get_input_element_type(0);
     } else if (node.get_output_size() > 0) {

diff --git a/modules/nvidia_plugin/src/cuda_operation_base.hpp b/modules/nvidia_plugin/src/cuda_operation_base.hpp
@@ -4,8 +4,6 @@
 
 #pragma once
 
-#include <ie_layouts.h>
-
 #include <cuda/device_pointers.hpp>
 #include <cuda_creation_context.hpp>
 #include <cuda_inference_request_context.hpp>

diff --git a/modules/nvidia_plugin/src/cuda_operation_registry.hpp b/modules/nvidia_plugin/src/cuda_operation_registry.hpp
@@ -10,6 +10,7 @@
 #include <unordered_map>
 
 #include "cuda_operation_base.hpp"
+#include "openvino/core/node.hpp"
 
 namespace ov {
 namespace nvidia_gpu {

diff --git a/modules/nvidia_plugin/src/cuda_plugin.cpp b/modules/nvidia_plugin/src/cuda_plugin.cpp
@@ -3,18 +3,15 @@
 //
 #include <fmt/format.h>
 
-#include "ie_metric_helpers.hpp"
-
-#include "cpp_interfaces/interface/ie_internal_plugin_config.hpp"
 #include "cuda/props.hpp"
 #include "cuda_compiled_model.hpp"
 #include "cuda_infer_request.hpp"
 #include "cuda_itt.hpp"
 #include "cuda_operation_registry.hpp"
 #include "cuda_plugin.hpp"
-#include "nvidia/nvidia_config.hpp"
 #include "openvino/core/op_extension.hpp"
 #include "openvino/op/util/op_types.hpp"
+#include "openvino/runtime/internal_properties.hpp"
 #include "openvino/runtime/core.hpp"
 #include "openvino/runtime/properties.hpp"
 #include "openvino/runtime/threading/executor_manager.hpp"
@@ -74,7 +71,8 @@ std::shared_ptr<ov::ICompiledModel> Plugin::compile_model(const std::shared_ptr<
     auto compiled_model = std::make_shared<CompiledModel>(model->clone(),
                                                           full_config,
                                                           wait_executor,
-                                                          shared_from_this());
+                                                          shared_from_this(),
+                                                          false);
     return compiled_model;
 }
 
@@ -105,13 +103,22 @@ std::shared_ptr<ov::ICompiledModel> Plugin::import_model(std::istream& model_str
 
     auto model = get_core()->read_model(xml_string, weights);
 
-    auto full_config = get_full_config(properties);
+    // check ov::loaded_from_cache property and erase it due to not needed any more.
+    auto _properties = properties;
+    const auto& it = _properties.find(ov::loaded_from_cache.name());
+    bool loaded_from_cache = false;
+    if (it != _properties.end()) {
+        loaded_from_cache = it->second.as<bool>();
+        _properties.erase(it);
+    }
+
+    auto full_config = get_full_config(_properties);
     auto wait_executor = get_stream_executor(full_config);
     auto compiled_model= std::make_shared<CompiledModel>(model,
                                                          full_config,
                                                          wait_executor,
                                                          shared_from_this(),
-                                                         true);
+                                                         loaded_from_cache);
     return compiled_model;
 }
 
@@ -193,8 +200,6 @@ void Plugin::set_property(const ov::AnyMap& properties) {
 }
 
 ov::Any Plugin::get_property(const std::string& name, const ov::AnyMap& properties) const {
-    using namespace InferenceEngine::CUDAMetrics;
-
     auto full_config = get_full_config(properties);
 
     if (ov::supported_properties == name) {
@@ -233,7 +238,7 @@ ov::Any Plugin::get_property(const std::string& name, const ov::AnyMap& properti
             ov::device::capability::EXPORT_IMPORT,
             ov::device::capability::FP32,
             ov::device::capability::FP16}};
-     } else if (ov::range_for_streams == name) {
+    } else if (ov::range_for_streams == name) {
         return decltype(ov::range_for_streams)::value_type{1, Configuration::reasonable_limit_of_streams};
     } else if (ov::range_for_async_infer_requests == name) {
         return decltype(ov::range_for_async_infer_requests)::value_type{1, 1, 1};

diff --git a/modules/nvidia_plugin/src/cuda_thread_pool.cpp b/modules/nvidia_plugin/src/cuda_thread_pool.cpp
@@ -6,8 +6,6 @@
 
 #include <fmt/format.h>
 
-#include <details/ie_exception.hpp>
-
 #include "cuda_latch.hpp"
 
 namespace ov {

diff --git a/modules/nvidia_plugin/src/kernels/details/error.cpp b/modules/nvidia_plugin/src/kernels/details/error.cpp
@@ -4,7 +4,6 @@
 
 #include <fmt/format.h>
 
-#include <details/ie_exception.hpp>
 #include <error.hpp>
 #include <exception>
 

diff --git a/modules/nvidia_plugin/src/memory_manager/cuda_device_mem_block.cpp b/modules/nvidia_plugin/src/memory_manager/cuda_device_mem_block.cpp
@@ -7,7 +7,6 @@
 #include <cuda_runtime_api.h>
 
 #include <cuda/runtime.hpp>
-#include <details/ie_exception.hpp>
 #include <iostream>
 
 namespace ov {

diff --git a/modules/nvidia_plugin/src/ops/convolution_components/convolution_cudnn_components.cpp b/modules/nvidia_plugin/src/ops/convolution_components/convolution_cudnn_components.cpp
@@ -9,7 +9,6 @@
 
 #include <cuda_config.hpp>
 #include <openvino/core/except.hpp>
-#include <nvidia/nvidia_config.hpp>
 #include <ops/converters.hpp>
 
 namespace ov::nvidia_gpu::Convolution::Details {

diff --git a/modules/nvidia_plugin/src/ops/convolution_cudnn.cpp b/modules/nvidia_plugin/src/ops/convolution_cudnn.cpp
@@ -6,7 +6,6 @@
 
 #include <cudnn.h>
 
-#include <details/ie_exception.hpp>
 #include <openvino/core/except.hpp>
 
 #include "cuda/constant_factory.hpp"

diff --git a/modules/nvidia_plugin/src/ops/convolution_cudnn_be.cpp b/modules/nvidia_plugin/src/ops/convolution_cudnn_be.cpp
@@ -7,7 +7,6 @@
 #include <fmt/format.h>
 
 #include <algorithm>
-#include <details/ie_exception.hpp>
 #include <openvino/core/except.hpp>
 #include <ops/converters.hpp>
 

diff --git a/modules/nvidia_plugin/src/ops/details/cuda_ie_api_import_fix.hpp b/modules/nvidia_plugin/src/ops/details/cuda_ie_api_import_fix.hpp
diff --git a/modules/nvidia_plugin/src/ops/details/cuda_ngraph_import_fix.hpp b/modules/nvidia_plugin/src/ops/details/cuda_ngraph_import_fix.hpp
diff --git a/modules/nvidia_plugin/src/ops/fused_convolution_cudnn.cpp b/modules/nvidia_plugin/src/ops/fused_convolution_cudnn.cpp
@@ -6,7 +6,6 @@
 
 #include <cudnn.h>
 
-#include <details/ie_exception.hpp>
 #include <openvino/core/except.hpp>
 #include <ops/converters.hpp>
 

diff --git a/modules/nvidia_plugin/src/ops/fused_convolution_cudnn_be.cpp b/modules/nvidia_plugin/src/ops/fused_convolution_cudnn_be.cpp
@@ -8,7 +8,6 @@
 
 #include <algorithm>
 #include <cuda/constant_factory.hpp>
-#include <details/ie_exception.hpp>
 #include <openvino/core/except.hpp>
 #include <ops/converters.hpp>
 

diff --git a/modules/nvidia_plugin/src/ops/fused_convolution_cudnn_decomposed.cpp b/modules/nvidia_plugin/src/ops/fused_convolution_cudnn_decomposed.cpp
@@ -6,7 +6,6 @@
 
 #include <cudnn.h>
 
-#include <details/ie_exception.hpp>
 #include <openvino/core/except.hpp>
 #include <ops/converters.hpp>
 

diff --git a/modules/nvidia_plugin/src/ops/logical_not.cpp b/modules/nvidia_plugin/src/ops/logical_not.cpp
@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: Apache-2.0
 //
 #include "logical_not.hpp"
+#include "openvino/core/except.hpp"
 
 #include <cuda_operation_registry.hpp>
 

diff --git a/modules/nvidia_plugin/src/transformer/cuda_graph_transformer.hpp b/modules/nvidia_plugin/src/transformer/cuda_graph_transformer.hpp
@@ -6,7 +6,6 @@
 
 #include <cuda/runtime.hpp>
 
-#include "cpp/ie_cnn_network.h"
 #include "cuda_config.hpp"
 #include "openvino/core/model.hpp"